Cordelia Schmid

Antoine Yang, Antoine Miech, Josef Sivic et al. · deepmind

Video question answering (VideoQA) is a complex task that requires diverse multi-modal data for training. Manual annotation of question and answers for videos, however, is tedious and prohibits scalability. To tackle this problem, recent methods consider zero-shot settings with no manual annotation of visual question-answer. In particular, a promising approach adapts frozen autoregressive language models pretrained on Web-scale text-only data to multi-modal inputs. In contrast, we here build on frozen bidirectional language models (BiLM) and show that such an approach provides a stronger and cheaper alternative for zero-shot VideoQA. In particular, (i) we combine visual inputs with the frozen BiLM using light trainable modules, (ii) we train such modules using Web-scraped multi-modal data, and finally (iii) we perform zero-shot VideoQA inference through masked language modeling, where the masked text is the answer to a given question. Our proposed approach, FrozenBiLM, outperforms the state of the art in zero-shot VideoQA by a significant margin on a variety of datasets, including LSMDC-FiB, iVQA, MSRVTT-QA, MSVD-QA, ActivityNet-QA, TGIF-FrameQA, How2QA and TVQA. It also demonstrates competitive performance in the few-shot and fully-supervised setting. Our code and models are publicly available at https://github.com/antoyang/FrozenBiLM.

Antoine Yang, Arsha Nagrani, Paul Hongsuck Seo et al. · deepmind

In this work, we introduce Vid2Seq, a multi-modal single-stage dense event captioning model pretrained on narrated videos which are readily-available at scale. The Vid2Seq architecture augments a language model with special time tokens, allowing it to seamlessly predict event boundaries and textual descriptions in the same output sequence. Such a unified model requires large-scale training data, which is not available in current annotated datasets. We show that it is possible to leverage unlabeled narrated videos for dense video captioning, by reformulating sentence boundaries of transcribed speech as pseudo event boundaries, and using the transcribed speech sentences as pseudo event captions. The resulting Vid2Seq model pretrained on the YT-Temporal-1B dataset improves the state of the art on a variety of dense video captioning benchmarks including YouCook2, ViTT and ActivityNet Captions. Vid2Seq also generalizes well to the tasks of video paragraph captioning and video clip captioning, and to few-shot settings. Our code is publicly available at https://antoyang.github.io/vid2seq.html.

Ziniu Hu, Ahmet Iscen, Chen Sun et al. · cmu

In this paper, we propose an end-to-end Retrieval-Augmented Visual Language Model (REVEAL) that learns to encode world knowledge into a large-scale memory, and to retrieve from it to answer knowledge-intensive queries. REVEAL consists of four key components: the memory, the encoder, the retriever and the generator. The large-scale memory encodes various sources of multimodal world knowledge (e.g. image-text pairs, question answering pairs, knowledge graph triplets, etc) via a unified encoder. The retriever finds the most relevant knowledge entries in the memory, and the generator fuses the retrieved knowledge with the input query to produce the output. A key novelty in our approach is that the memory, encoder, retriever and generator are all pre-trained end-to-end on a massive amount of data. Furthermore, our approach can use a diverse set of multimodal knowledge sources, which is shown to result in significant gains. We show that REVEAL achieves state-of-the-art results on visual question answering and image captioning.

Karsten Roth, Jae Myung Kim, A. Sophia Koepke et al.

The visual classification performance of vision-language models such as CLIP has been shown to benefit from additional semantic knowledge from large language models (LLMs) such as GPT-3. In particular, averaging over LLM-generated class descriptors, e.g. "waffle, which has a round shape", can notably improve generalization performance. In this work, we critically study this behavior and propose WaffleCLIP, a framework for zero-shot visual classification which simply replaces LLM-generated descriptors with random character and word descriptors. Without querying external models, we achieve comparable performance gains on a large number of visual classification tasks. This allows WaffleCLIP to both serve as a low-cost alternative, as well as a sanity check for any future LLM-based vision-language model extensions. We conduct an extensive experimental study on the impact and shortcomings of additional semantics introduced with LLM-generated descriptors, and showcase how - if available - semantic context is better leveraged by querying LLMs for high-level concepts, which we show can be done to jointly resolve potential class name ambiguities. Code is available here: https://github.com/ExplainableML/WaffleCLIP.

Shen Yan, Xuehan Xiong, Arsha Nagrani et al.

While large-scale image-text pretrained models such as CLIP have been used for multiple video-level tasks on trimmed videos, their use for temporal localization in untrimmed videos is still a relatively unexplored task. We design a new approach for this called UnLoc, which uses pretrained image and text towers, and feeds tokens to a video-text fusion model. The output of the fusion module are then used to construct a feature pyramid in which each level connects to a head to predict a per-frame relevancy score and start/end time displacements. Unlike previous works, our architecture enables Moment Retrieval, Temporal Localization, and Action Segmentation with a single stage model, without the need for action proposals, motion based pretrained features or representation masking. Unlike specialized models, we achieve state of the art results on all three different localization tasks with a unified approach. Code will be available at: \url{https://github.com/google-research/scenic}.

Sanjay Subramanian, Medhini Narasimhan, Kushal Khangaonkar et al. · berkeley

We present a framework that formulates visual question answering as modular code generation. In contrast to prior work on modular approaches to VQA, our approach requires no additional training and relies on pre-trained language models (LMs), visual models pre-trained on image-caption pairs, and fifty VQA examples used for in-context learning. The generated Python programs invoke and compose the outputs of the visual models using arithmetic and conditional logic. Our approach improves accuracy on the COVR dataset by at least 3% and on the GQA dataset by roughly 2% compared to the few-shot baseline that does not employ code generation.

Jae Myung Kim, A. Sophia Koepke, Cordelia Schmid et al.

Cross-modal retrieval methods are the preferred tool to search databases for the text that best matches a query image and vice versa. However, image-text retrieval models commonly learn to memorize spurious correlations in the training data, such as frequent object co-occurrence, instead of looking at the actual underlying reasons for the prediction in the image. For image-text retrieval, this manifests in retrieved sentences that mention objects that are not present in the query image. In this work, we introduce ODmAP@k, an object decorrelation metric that measures a model's robustness to spurious correlations in the training data. We use automatic image and text manipulations to control the presence of such object correlations in designated test data. Additionally, our data synthesis technique is used to tackle model biases due to spurious correlations of semantically unrelated objects in the training data. We apply our proposed pipeline, which involves the finetuning of image-text retrieval frameworks on carefully designed synthetic data, to three state-of-the-art models for image-text retrieval. This results in significant improvements for all three models, both in terms of the standard retrieval performance and in terms of our object decorrelation metric. The code is available at https://github.com/ExplainableML/Spurious_CM_Retrieval.

Lucas Ventura, Antoine Yang, Cordelia Schmid et al. · deepmind

Composed Image Retrieval (CoIR) has recently gained popularity as a task that considers both text and image queries together, to search for relevant images in a database. Most CoIR approaches require manually annotated datasets, comprising image-text-image triplets, where the text describes a modification from the query image to the target image. However, manual curation of CoIR triplets is expensive and prevents scalability. In this work, we instead propose a scalable automatic dataset creation methodology that generates triplets given video-caption pairs, while also expanding the scope of the task to include composed video retrieval (CoVR). To this end, we mine paired videos with a similar caption from a large database, and leverage a large language model to generate the corresponding modification text. Applying this methodology to the extensive WebVid2M collection, we automatically construct our WebVid-CoVR dataset, resulting in 1.6 million triplets. Moreover, we introduce a new benchmark for CoVR with a manually annotated evaluation set, along with baseline results. We further validate that our methodology is equally applicable to image-caption pairs, by generating 3.3 million CoIR training triplets using the Conceptual Captions dataset. Our model builds on BLIP-2 pretraining, adapting it to composed video (or image) retrieval, and incorporates an additional caption retrieval loss to exploit extra supervision beyond the triplet. We provide extensive ablations to analyze the design choices on our new CoVR benchmark. Our experiments also demonstrate that training a CoVR model on our datasets effectively transfers to CoIR, leading to improved state-of-the-art performance in the zero-shot setup on the CIRR, FashionIQ, and CIRCO benchmarks. Our code, datasets, and models are publicly available at https://imagine.enpc.fr/~ventural/covr/.

Antoine Yang, Antoine Miech, Josef Sivic et al. · deepmind

Recent methods for visual question answering rely on large-scale annotated datasets. Manual annotation of questions and answers for videos, however, is tedious, expensive and prevents scalability. In this work, we propose to avoid manual annotation and generate a large-scale training dataset for video question answering making use of automatic cross-modal supervision. We leverage a question generation transformer trained on text data and use it to generate question-answer pairs from transcribed video narrations. Given narrated videos, we then automatically generate the HowToVQA69M dataset with 69M video-question-answer triplets. To handle the open vocabulary of diverse answers in this dataset, we propose a training procedure based on a contrastive loss between a video-question multi-modal transformer and an answer transformer. We introduce the zero-shot VideoQA task and the VideoQA feature probe evaluation setting and show excellent results, in particular for rare answers. Furthermore, our method achieves competitive results on MSRVTT-QA, ActivityNet-QA, MSVD-QA and How2QA datasets. We also show that our VideoQA dataset generation approach generalizes to another source of web video and text data. We use our method to generate the WebVidVQA3M dataset from the WebVid dataset, i.e., videos with alt-text annotations, and show its benefits for training VideoQA models. Finally, for a detailed evaluation we introduce iVQA, a new VideoQA dataset with reduced language bias and high-quality manual annotations. Code, datasets and trained models are available at https://antoyang.github.io/just-ask.html

Antoine Yang, Arsha Nagrani, Ivan Laptev et al. · deepmind

Segmenting long videos into chapters enables users to quickly navigate to the information of their interest. This important topic has been understudied due to the lack of publicly released datasets. To address this issue, we present VidChapters-7M, a dataset of 817K user-chaptered videos including 7M chapters in total. VidChapters-7M is automatically created from videos online in a scalable manner by scraping user-annotated chapters and hence without any additional manual annotation. We introduce the following three tasks based on this data. First, the video chapter generation task consists of temporally segmenting the video and generating a chapter title for each segment. To further dissect the problem, we also define two variants of this task: video chapter generation given ground-truth boundaries, which requires generating a chapter title given an annotated video segment, and video chapter grounding, which requires temporally localizing a chapter given its annotated title. We benchmark both simple baselines and state-of-the-art video-language models for these three tasks. We also show that pretraining on VidChapters-7M transfers well to dense video captioning tasks in both zero-shot and finetuning settings, largely improving the state of the art on the YouCook2 and ViTT benchmarks. Finally, our experiments reveal that downstream performance scales well with the size of the pretraining dataset. Our dataset, code, and models are publicly available at https://antoyang.github.io/vidchapters.html.

Youngwook Kim, Jae Myung Kim, Jieun Jeong et al.

Due to the expensive costs of collecting labels in multi-label classification datasets, partially annotated multi-label classification has become an emerging field in computer vision. One baseline approach to this task is to assume unobserved labels as negative labels, but this assumption induces label noise as a form of false negative. To understand the negative impact caused by false negative labels, we study how these labels affect the model's explanation. We observe that the explanation of two models, trained with full and partial labels each, highlights similar regions but with different scaling, where the latter tends to have lower attribution scores. Based on these findings, we propose to boost the attribution scores of the model trained with partial labels to make its explanation resemble that of the model trained with full labels. Even with the conceptually simple approach, the multi-label classification performance improves by a large margin in three different datasets on a single positive label setting and one on a large-scale partial label setting. Code is available at https://github.com/youngwk/BridgeGapExplanationPAMC.

Xingyi Zhou, Anurag Arnab, Chen Sun et al.

We propose a new task and model for dense video object captioning -- detecting, tracking and captioning trajectories of objects in a video. This task unifies spatial and temporal localization in video, whilst also requiring fine-grained visual understanding that is best described by natural language. We propose a unified model, and demonstrate how our end-to-end approach is more accurate and temporally coherent than a multi-stage pipeline combining state-of-the-art detection, tracking, and captioning models. Moreover, we propose a training strategy based on a mixture of disjoint tasks, which allows us to leverage diverse, large-scale datasets which supervise different parts of our model. Although each pretraining task only provides weak supervision, they are complementary and, when combined, result in noteworthy zero-shot ability and serve as strong initialization for additional finetuning to further improve accuracy. We carefully design new metrics capturing all components of our task, and show how we can repurpose existing video grounding datasets (e.g. VidSTG and VLN) for our new task. We show that our model improves upon a number of strong baselines for this new task. Furthermore, we can apply our model to the task of spatial grounding, outperforming prior state-of-the-art on VidSTG and VLN, without explicitly training for it. Code is available at https://github.com/google-research/scenic/tree/main/scenic/projects/densevoc.

Pierre-Louis Guhur, Shizhe Chen, Ricardo Garcia et al.

In human environments, robots are expected to accomplish a variety of manipulation tasks given simple natural language instructions. Yet, robotic manipulation is extremely challenging as it requires fine-grained motor control, long-term memory as well as generalization to previously unseen tasks and environments. To address these challenges, we propose a unified transformer-based approach that takes into account multiple inputs. In particular, our transformer architecture integrates (i) natural language instructions and (ii) multi-view scene observations while (iii) keeping track of the full history of observations and actions. Such an approach enables learning dependencies between history and instructions and improves manipulation precision using multiple views. We evaluate our method on the challenging RLBench benchmark and on a real-world robot. Notably, our approach scales to 74 diverse RLBench tasks and outperforms the state of the art. We also address instruction-conditioned tasks and demonstrate excellent generalization to previously unseen variations.

Shizhe Chen, Pierre-Louis Guhur, Makarand Tapaswi et al.

Localizing objects in 3D scenes based on natural language requires understanding and reasoning about spatial relations. In particular, it is often crucial to distinguish similar objects referred by the text, such as "the left most chair" and "a chair next to the window". In this work we propose a language-conditioned transformer model for grounding 3D objects and their spatial relations. To this end, we design a spatial self-attention layer that accounts for relative distances and orientations between objects in input 3D point clouds. Training such a layer with visual and language inputs enables to disambiguate spatial relations and to localize objects referred by the text. To facilitate the cross-modal learning of relations, we further propose a teacher-student approach where the teacher model is first trained using ground-truth object labels, and then helps to train a student model using point cloud inputs. We perform ablation studies showing advantages of our approach. We also demonstrate our model to significantly outperform the state of the art on the challenging Nr3D, Sr3D and ScanRefer 3D object grounding datasets.

Arsha Nagrani, Paul Hongsuck Seo, Bryan Seybold et al.

A major challenge in text-video and text-audio retrieval is the lack of large-scale training data. This is unlike image-captioning, where datasets are in the order of millions of samples. To close this gap we propose a new video mining pipeline which involves transferring captions from image captioning datasets to video clips with no additional manual effort. Using this pipeline, we create a new large-scale, weakly labelled audio-video captioning dataset consisting of millions of paired clips and captions. We show that training a multimodal transformed based model on this data achieves competitive performance on video retrieval and video captioning, matching or even outperforming HowTo100M pretraining with 20x fewer clips. We also show that our mined clips are suitable for text-audio pretraining, and achieve state of the art results for the task of audio retrieval.

Liliane Momeni, Mathilde Caron, Arsha Nagrani et al.

Understanding verbs is crucial to modelling how people and objects interact with each other and the environment through space and time. Recently, state-of-the-art video-language models based on CLIP have been shown to have limited verb understanding and to rely extensively on nouns, restricting their performance in real-world video applications that require action and temporal understanding. In this work, we improve verb understanding for CLIP-based video-language models by proposing a new Verb-Focused Contrastive (VFC) framework. This consists of two main components: (1) leveraging pretrained large language models (LLMs) to create hard negatives for cross-modal contrastive learning, together with a calibration strategy to balance the occurrence of concepts in positive and negative pairs; and (2) enforcing a fine-grained, verb phrase alignment loss. Our method achieves state-of-the-art results for zero-shot performance on three downstream tasks that focus on verb understanding: video-text matching, video question-answering and video classification. To the best of our knowledge, this is the first work which proposes a method to alleviate the verb understanding problem, and does not simply highlight it.

Zerui Chen, Yana Hasson, Cordelia Schmid et al.

Recent work achieved impressive progress towards joint reconstruction of hands and manipulated objects from monocular color images. Existing methods focus on two alternative representations in terms of either parametric meshes or signed distance fields (SDFs). On one side, parametric models can benefit from prior knowledge at the cost of limited shape deformations and mesh resolutions. Mesh models, hence, may fail to precisely reconstruct details such as contact surfaces of hands and objects. SDF-based methods, on the other side, can represent arbitrary details but are lacking explicit priors. In this work we aim to improve SDF models using priors provided by parametric representations. In particular, we propose a joint learning framework that disentangles the pose and the shape. We obtain hand and object poses from parametric models and use them to align SDFs in 3D space. We show that such aligned SDFs better focus on reconstructing shape details and improve reconstruction accuracy both for hands and objects. We evaluate our method and demonstrate significant improvements over the state of the art on the challenging ObMan and DexYCB benchmarks.

Zerui Chen, Shizhe Chen, Cordelia Schmid et al.

Signed distance functions (SDFs) is an attractive framework that has recently shown promising results for 3D shape reconstruction from images. SDFs seamlessly generalize to different shape resolutions and topologies but lack explicit modelling of the underlying 3D geometry. In this work, we exploit the hand structure and use it as guidance for SDF-based shape reconstruction. In particular, we address reconstruction of hands and manipulated objects from monocular RGB images. To this end, we estimate poses of hands and objects and use them to guide 3D reconstruction. More specifically, we predict kinematic chains of pose transformations and align SDFs with highly-articulated hand poses. We improve the visual features of 3D points with geometry alignment and further leverage temporal information to enhance the robustness to occlusion and motion blurs. We conduct extensive experiments on the challenging ObMan and DexYCB benchmarks and demonstrate significant improvements of the proposed method over the state of the art.

Shizhe Chen, Pierre-Louis Guhur, Makarand Tapaswi et al.

In vision-and-language navigation (VLN), an embodied agent is required to navigate in realistic 3D environments following natural language instructions. One major bottleneck for existing VLN approaches is the lack of sufficient training data, resulting in unsatisfactory generalization to unseen environments. While VLN data is typically collected manually, such an approach is expensive and prevents scalability. In this work, we address the data scarcity issue by proposing to automatically create a large-scale VLN dataset from 900 unlabeled 3D buildings from HM3D. We generate a navigation graph for each building and transfer object predictions from 2D to generate pseudo 3D object labels by cross-view consistency. We then fine-tune a pretrained language model using pseudo object labels as prompts to alleviate the cross-modal gap in instruction generation. Our resulting HM3D-AutoVLN dataset is an order of magnitude larger than existing VLN datasets in terms of navigation environments and instructions. We experimentally demonstrate that HM3D-AutoVLN significantly increases the generalization ability of resulting VLN models. On the SPL metric, our approach improves over state of the art by 7.1% and 8.1% on the unseen validation splits of REVERIE and SOON datasets respectively.

Mariana-Iuliana Georgescu, Eduardo Fonseca, Radu Tudor Ionescu et al.

Can we leverage the audiovisual information already present in video to improve self-supervised representation learning? To answer this question, we study various pretraining architectures and objectives within the masked autoencoding framework, motivated by the success of similar methods in natural language and image understanding. We show that we can achieve significant improvements on audiovisual downstream classification tasks, surpassing the state-of-the-art on VGGSound and AudioSet. Furthermore, we can leverage our audiovisual pretraining scheme for multiple unimodal downstream tasks using a single audiovisual pretrained model. We additionally demonstrate the transferability of our representations, achieving state-of-the-art audiovisual results on Epic Kitchens without pretraining specifically for this dataset.

Jae Myung Kim, Jessica Bader, Stephan Alaniz et al.

While text-to-image diffusion models have been shown to achieve state-of-the-art results in image synthesis, they have yet to prove their effectiveness in downstream applications. Previous work has proposed to generate data for image classifier training given limited real data access. However, these methods struggle to generate in-distribution images or depict fine-grained features, thereby hindering the generalization of classification models trained on synthetic datasets. We propose DataDream, a framework for synthesizing classification datasets that more faithfully represents the real data distribution when guided by few-shot examples of the target classes. DataDream fine-tunes LoRA weights for the image generation model on the few real images before generating the training data using the adapted model. We then fine-tune LoRA weights for CLIP using the synthetic data to improve downstream image classification over previous approaches on a large variety of datasets. We demonstrate the efficacy of DataDream through extensive experiments, surpassing state-of-the-art classification accuracy with few-shot data across 7 out of 10 datasets, while being competitive on the other 3. Additionally, we provide insights into the impact of various factors, such as the number of real-shot and generated images as well as the fine-tuning compute on model performance. The code is available at https://github.com/ExplainableML/DataDream.

Matthieu Futeral, Cordelia Schmid, Ivan Laptev et al.

One of the major challenges of machine translation (MT) is ambiguity, which can in some cases be resolved by accompanying context such as images. However, recent work in multimodal MT (MMT) has shown that obtaining improvements from images is challenging, limited not only by the difficulty of building effective cross-modal representations, but also by the lack of specific evaluation and training data. We present a new MMT approach based on a strong text-only MT model, which uses neural adapters, a novel guided self-attention mechanism and which is jointly trained on both visually-conditioned masking and MMT. We also introduce CoMMuTE, a Contrastive Multilingual Multimodal Translation Evaluation set of ambiguous sentences and their possible translations, accompanied by disambiguating images corresponding to each translation. Our approach obtains competitive results compared to strong text-only models on standard English-to-French, English-to-German and English-to-Czech benchmarks and outperforms baselines and state-of-the-art MMT systems by a large margin on our contrastive test set. Our code and CoMMuTE are freely available.

Medhini Narasimhan, Arsha Nagrani, Chen Sun et al.

YouTube users looking for instructions for a specific task may spend a long time browsing content trying to find the right video that matches their needs. Creating a visual summary (abridged version of a video) provides viewers with a quick overview and massively reduces search time. In this work, we focus on summarizing instructional videos, an under-explored area of video summarization. In comparison to generic videos, instructional videos can be parsed into semantically meaningful segments that correspond to important steps of the demonstrated task. Existing video summarization datasets rely on manual frame-level annotations, making them subjective and limited in size. To overcome this, we first automatically generate pseudo summaries for a corpus of instructional videos by exploiting two key assumptions: (i) relevant steps are likely to appear in multiple videos of the same task (Task Relevance), and (ii) they are more likely to be described by the demonstrator verbally (Cross-Modal Saliency). We propose an instructional video summarization network that combines a context-aware temporal video encoder and a segment scoring transformer. Using pseudo summaries as weak supervision, our network constructs a visual summary for an instructional video given only video and transcribed speech. To evaluate our model, we collect a high-quality test set, WikiHow Summaries, by scraping WikiHow articles that contain video demonstrations and visual depictions of steps allowing us to obtain the ground-truth summaries. We outperform several baselines and a state-of-the-art video summarization model on this new benchmark.

Ahmet Iscen, Mathilde Caron, Alireza Fathi et al.

Contrastive image-text models such as CLIP form the building blocks of many state-of-the-art systems. While they excel at recognizing common generic concepts, they still struggle on fine-grained entities which are rare, or even absent from the pre-training dataset. Hence, a key ingredient to their success has been the use of large-scale curated pre-training data aiming at expanding the set of concepts that they can memorize during the pre-training stage. In this work, we explore an alternative to encoding fine-grained knowledge directly into the model's parameters: we instead train the model to retrieve this knowledge from an external memory. Specifically, we propose to equip existing vision-text models with the ability to refine their embedding with cross-modal retrieved information from a memory at inference time, which greatly improves their zero-shot predictions. Remarkably, we show that this can be done with a light-weight, single-layer, fusion transformer on top of a frozen CLIP. Our experiments validate that our retrieval-enhanced contrastive (RECO) training improves CLIP performance substantially on several challenging fine-grained tasks: for example +10.9 on Stanford Cars, +10.2 on CUB-2011 and +7.3 on the recent OVEN benchmark, where we even outperform the fine-tuned models on unseen classes.

Shizhe Chen, Ricardo Garcia, Cordelia Schmid et al.

The ability for robots to comprehend and execute manipulation tasks based on natural language instructions is a long-term goal in robotics. The dominant approaches for language-guided manipulation use 2D image representations, which face difficulties in combining multi-view cameras and inferring precise 3D positions and relationships. To address these limitations, we propose a 3D point cloud based policy called PolarNet for language-guided manipulation. It leverages carefully designed point cloud inputs, efficient point cloud encoders, and multimodal transformers to learn 3D point cloud representations and integrate them with language instructions for action prediction. PolarNet is shown to be effective and data efficient in a variety of experiments conducted on the RLBench benchmark. It outperforms state-of-the-art 2D and 3D approaches in both single-task and multi-task learning. It also achieves promising results on a real robot.

Minttu Alakuijala, Gabriel Dulac-Arnold, Julien Mairal et al.

Observing a human demonstrator manipulate objects provides a rich, scalable and inexpensive source of data for learning robotic policies. However, transferring skills from human videos to a robotic manipulator poses several challenges, not least a difference in action and observation spaces. In this work, we use unlabeled videos of humans solving a wide range of manipulation tasks to learn a task-agnostic reward function for robotic manipulation policies. Thanks to the diversity of this training data, the learned reward function sufficiently generalizes to image observations from a previously unseen robot embodiment and environment to provide a meaningful prior for directed exploration in reinforcement learning. We propose two methods for scoring states relative to a goal image: through direct temporal regression, and through distances in an embedding space obtained with time-contrastive learning. By conditioning the function on a goal image, we are able to reuse one model across a variety of tasks. Unlike prior work on leveraging human videos to teach robots, our method, Human Offline Learned Distances (HOLD) requires neither a priori data from the robot environment, nor a set of task-specific human demonstrations, nor a predefined notion of correspondence across morphologies, yet it is able to accelerate training of several manipulation tasks on a simulated robot arm compared to using only a sparse reward obtained from task completion.

Shizhe Chen, Thomas Chabal, Ivan Laptev et al.

Object goal navigation aims to navigate an agent to locations of a given object category in unseen environments. Classical methods explicitly build maps of environments and require extensive engineering while lacking semantic information for object-oriented exploration. On the other hand, end-to-end learning methods alleviate manual map design and predict actions using implicit representations. Such methods, however, lack an explicit notion of geometry and may have limited ability to encode navigation history. In this work, we propose an implicit spatial map for object goal navigation. Our implicit map is recursively updated with new observations at each step using a transformer. To encourage spatial reasoning, we introduce auxiliary tasks and train our model to reconstruct explicit maps as well as to predict visual features, semantic labels and actions. Our method significantly outperforms the state of the art on the challenging MP3D dataset and generalizes well to the HM3D dataset. We successfully deploy our model on a real robot and achieve encouraging object goal navigation results in real scenes using only a few real-world demonstrations. Code, trained models and videos are available at \url{https://www.di.ens.fr/willow/research/onav_rim/}.

Ahmet Iscen, Alireza Fathi, Cordelia Schmid

Retrieval augmented models are becoming increasingly popular for computer vision tasks after their recent success in NLP problems. The goal is to enhance the recognition capabilities of the model by retrieving similar examples for the visual input from an external memory set. In this work, we introduce an attention-based memory module, which learns the importance of each retrieved example from the memory. Compared to existing approaches, our method removes the influence of the irrelevant retrieved examples, and retains those that are beneficial to the input query. We also thoroughly study various ways of constructing the memory dataset. Our experiments show the benefit of using a massive-scale memory dataset of 1B image-text pairs, and demonstrate the performance of different memory representations. We evaluate our method in three different classification tasks, namely long-tailed recognition, learning with noisy labels, and fine-grained classification, and show that it achieves state-of-the-art accuracies in ImageNet-LT, Places-LT and Webvision datasets.

Robin Strudel, Ivan Laptev, Cordelia Schmid

Visual grounding localizes regions (boxes or segments) in the image corresponding to given referring expressions. In this work we address image segmentation from referring expressions, a problem that has so far only been addressed in a fully-supervised setting. A fully-supervised setup, however, requires pixel-wise supervision and is hard to scale given the expense of manual annotation. We therefore introduce a new task of weakly-supervised image segmentation from referring expressions and propose Text grounded semantic SEGgmentation (TSEG) that learns segmentation masks directly from image-level referring expressions without pixel-level annotations. Our transformer-based method computes patch-text similarities and guides the classification objective during training with a new multi-label patch assignment mechanism. The resulting visual grounding model segments image regions corresponding to given natural language expressions. Our approach TSEG demonstrates promising results for weakly-supervised referring expression segmentation on the challenging PhraseCut and RefCOCO datasets. TSEG also shows competitive performance when evaluated in a zero-shot setting for semantic segmentation on Pascal VOC.

Xingyi Zhou, Anurag Arnab, Chen Sun et al.

Current state-of-the-art video models process a video clip as a long sequence of spatio-temporal tokens. However, they do not explicitly model objects, their interactions across the video, and instead process all the tokens in the video. In this paper, we investigate how we can use knowledge of objects to design better video models, namely to process fewer tokens and to improve recognition accuracy. This is in contrast to prior works which either drop tokens at the cost of accuracy, or increase accuracy whilst also increasing the computation required. First, we propose an object-guided token sampling strategy that enables us to retain a small fraction of the input tokens with minimal impact on accuracy. And second, we propose an object-aware attention module that enriches our feature representation with object information and improves overall accuracy. Our resulting framework achieves better performance when using fewer tokens than strong baselines. In particular, we match our baseline with 30%, 40%, and 60% of the input tokens on SomethingElse, Something-something v2, and Epic-Kitchens, respectively. When we use our model to process the same number of tokens as our baseline, we improve by 0.6 to 4.2 points on these datasets.

Paul Hongsuck Seo, Arsha Nagrani, Cordelia Schmid

Audiovisual automatic speech recognition (AV-ASR) aims to improve the robustness of a speech recognition system by incorporating visual information. Training fully supervised multimodal models for this task from scratch, however is limited by the need for large labelled audiovisual datasets (in each downstream domain of interest). We present AVFormer, a simple method for augmenting audio-only models with visual information, at the same time performing lightweight domain adaptation. We do this by (i) injecting visual embeddings into a frozen ASR model using lightweight trainable adaptors. We show that these can be trained on a small amount of weakly labelled video data with minimum additional training time and parameters. (ii) We also introduce a simple curriculum scheme during training which we show is crucial to enable the model to jointly process audio and visual information effectively; and finally (iii) we show that our model achieves state of the art zero-shot results on three different AV-ASR benchmarks (How2, VisSpeech and Ego4D), while also crucially preserving decent performance on traditional audio-only speech recognition benchmarks (LibriSpeech). Qualitative results show that our model effectively leverages visual information for robust speech recognition.

Mathilde Caron, Neil Houlsby, Cordelia Schmid

Pixel-level labels are particularly expensive to acquire. Hence, pretraining is a critical step to improve models on a task like semantic segmentation. However, prominent algorithms for pretraining neural networks use image-level objectives, e.g. image classification, image-text alignment a la CLIP, or self-supervised contrastive learning. These objectives do not model spatial information, which might be sub-optimal when finetuning on downstream tasks with spatial reasoning. In this work, we pretrain network with a location-aware (LOCA) self-supervised method which fosters the emergence of strong dense features. Specifically, we use both a patch-level clustering scheme to mine dense pseudo-labels and a relative location prediction task to encourage learning about object parts and their spatial arrangements. Our experiments show that LOCA pretraining leads to representations that transfer competitively to challenging and diverse semantic segmentation datasets.

Alexey Gritsenko, Xuehan Xiong, Josip Djolonga et al.

The most performant spatio-temporal action localisation models use external person proposals and complex external memory banks. We propose a fully end-to-end, purely-transformer based model that directly ingests an input video, and outputs tubelets -- a sequence of bounding boxes and the action classes at each frame. Our flexible model can be trained with either sparse bounding-box supervision on individual frames, or full tubelet annotations. And in both cases, it predicts coherent tubelets as the output. Moreover, our end-to-end model requires no additional pre-processing in the form of proposals, or post-processing in terms of non-maximal suppression. We perform extensive ablation experiments, and significantly advance the state-of-the-art results on four different spatio-temporal action localisation benchmarks with both sparse keyframes and full tubelet annotations.

Xuehan Xiong, Anurag Arnab, Arsha Nagrani et al.

This report describes the approach behind our winning solution to the 2022 Epic-Kitchens Action Recognition Challenge. Our approach builds upon our recent work, Multiview Transformer for Video Recognition (MTV), and adapts it to multimodal inputs. Our final submission consists of an ensemble of Multimodal MTV (M&M) models varying backbone sizes and input modalities. Our approach achieved 52.8% Top-1 accuracy on the test set in action classes, which is 4.1% higher than last year's winning entry.

Sai Kumar Dwivedi, Cordelia Schmid, Hongwei Yi et al.

The regression of 3D Human Pose and Shape (HPS) from an image is becoming increasingly accurate. This makes the results useful for downstream tasks like human action recognition or 3D graphics. Yet, no regressor is perfect, and accuracy can be affected by ambiguous image evidence or by poses and appearance that are unseen during training. Most current HPS regressors, however, do not report the confidence of their outputs, meaning that downstream tasks cannot differentiate accurate estimates from inaccurate ones. To address this, we develop POCO, a novel framework for training HPS regressors to estimate not only a 3D human body, but also their confidence, in a single feed-forward pass. Specifically, POCO estimates both the 3D body pose and a per-sample variance. The key idea is to introduce a Dual Conditioning Strategy (DCS) for regressing uncertainty that is highly correlated to pose reconstruction quality. The POCO framework can be applied to any HPS regressor and here we evaluate it by modifying HMR, PARE, and CLIFF. In all cases, training the network to reason about uncertainty helps it learn to more accurately estimate 3D pose. While this was not our goal, the improvement is modest but consistent. Our main motivation is to provide uncertainty estimates for downstream tasks; we demonstrate this in two ways: (1) We use the confidence estimates to bootstrap HPS training. Given unlabelled image data, we take the confident estimates of a POCO-trained regressor as pseudo ground truth. Retraining with this automatically-curated data improves accuracy. (2) We exploit uncertainty in video pose estimation by automatically identifying uncertain frames (e.g. due to occlusion) and inpainting these from confident frames. Code and models will be available for research at https://poco.is.tue.mpg.de.

Valentin Gabeur, Paul Hongsuck Seo, Arsha Nagrani et al.

Audio-visual automatic speech recognition (AV-ASR) is an extension of ASR that incorporates visual cues, often from the movements of a speaker's mouth. Unlike works that simply focus on the lip motion, we investigate the contribution of entire visual frames (visual actions, objects, background etc.). This is particularly useful for unconstrained videos, where the speaker is not necessarily visible. To solve this task, we propose a new sequence-to-sequence AudioVisual ASR TrAnsformeR (AVATAR) which is trained end-to-end from spectrograms and full-frame RGB. To prevent the audio stream from dominating training, we propose different word-masking strategies, thereby encouraging our model to pay attention to the visual stream. We demonstrate the contribution of the visual modality on the How2 AV-ASR benchmark, especially in the presence of simulated noise, and show that our model outperforms all other prior work by a large margin. Finally, we also create a new, real-world test bed for AV-ASR called VisSpeech, which demonstrates the contribution of the visual modality under challenging audio conditions.

Ahmet Iscen, Thomas Bird, Mathilde Caron et al.

We study class-incremental learning, a training setup in which new classes of data are observed over time for the model to learn from. Despite the straightforward problem formulation, the naive application of classification models to class-incremental learning results in the "catastrophic forgetting" of previously seen classes. One of the most successful existing methods has been the use of a memory of exemplars, which overcomes the issue of catastrophic forgetting by saving a subset of past data into a memory bank and utilizing it to prevent forgetting when training future tasks. In our paper, we propose to enhance the utilization of this memory bank: we not only use it as a source of additional training data like existing works but also integrate it in the prediction process explicitly.Our method, the Memory Transformer Network (MTN), learns how to combine and aggregate the information from the nearest neighbors in the memory with a transformer to make more accurate predictions. We conduct extensive experiments and ablations to evaluate our approach. We show that MTN achieves state-of-the-art performance on the challenging ImageNet-1k and Google-Landmarks-1k incremental learning benchmarks.

Darshan Singh, Arsha Nagrani, Kawshik Manikantan et al.

Recent advancements in video models have shown tremendous progress, particularly in long video understanding. However, current benchmarks predominantly feature western-centric data and English as the dominant language, introducing significant biases in evaluation. To address this, we introduce CURVE (Cultural Understanding and Reasoning in Video Evaluation), a challenging benchmark for multicultural and multilingual video reasoning. CURVE comprises high-quality, entirely human-generated annotations from diverse, region-specific cultural videos across 18 global locales. Unlike prior work that relies on automatic translations, CURVE provides complex questions, answers, and multi-step reasoning steps, all crafted in native languages. Making progress on CURVE requires a deeply situated understanding of visual cultural context. Furthermore, we leverage CURVE's reasoning traces to construct evidence-based graphs and propose a novel iterative strategy using these graphs to identify fine-grained errors in reasoning. Our evaluations reveal that SoTA Video-LLMs struggle significantly, performing substantially below human-level accuracy, with errors primarily stemming from the visual perception of cultural elements. CURVE will be publicly available under https://github.com/google-deepmind/neptune?tab=readme-ov-file\#minerva-cultural

Ziniu Hu, Ahmet Iscen, Chen Sun et al.

In this paper, we propose an autonomous information seeking visual question answering framework, AVIS. Our method leverages a Large Language Model (LLM) to dynamically strategize the utilization of external tools and to investigate their outputs, thereby acquiring the indispensable knowledge needed to provide answers to the posed questions. Responding to visual questions that necessitate external knowledge, such as "What event is commemorated by the building depicted in this image?", is a complex task. This task presents a combinatorial search space that demands a sequence of actions, including invoking APIs, analyzing their responses, and making informed decisions. We conduct a user study to collect a variety of instances of human decision-making when faced with this task. This data is then used to design a system comprised of three components: an LLM-powered planner that dynamically determines which tool to use next, an LLM-powered reasoner that analyzes and extracts key information from the tool outputs, and a working memory component that retains the acquired information throughout the process. The collected user behavior serves as a guide for our system in two key ways. First, we create a transition graph by analyzing the sequence of decisions made by users. This graph delineates distinct states and confines the set of actions available at each state. Second, we use examples of user decision-making to provide our LLM-powered planner and reasoner with relevant contextual instances, enhancing their capacity to make informed decisions. We show that AVIS achieves state-of-the-art results on knowledge-intensive visual question answering benchmarks such as Infoseek and OK-VQA.

Anurag Arnab, Xuehan Xiong, Alexey Gritsenko et al.

Transfer learning is the predominant paradigm for training deep networks on small target datasets. Models are typically pretrained on large ``upstream'' datasets for classification, as such labels are easy to collect, and then finetuned on ``downstream'' tasks such as action localisation, which are smaller due to their finer-grained annotations. In this paper, we question this approach, and propose co-finetuning -- simultaneously training a single model on multiple ``upstream'' and ``downstream'' tasks. We demonstrate that co-finetuning outperforms traditional transfer learning when using the same total amount of data, and also show how we can easily extend our approach to multiple ``upstream'' datasets to further improve performance. In particular, co-finetuning significantly improves the performance on rare classes in our downstream task, as it has a regularising effect, and enables the network to learn feature representations that transfer between different datasets. Finally, we observe how co-finetuning with public, video classification datasets, we are able to achieve state-of-the-art results for spatio-temporal action localisation on the challenging AVA and AVA-Kinetics datasets, outperforming recent works which develop intricate models.

Ricardo Garcia, Robin Strudel, Shizhe Chen et al.

Learning visuomotor policies in simulation is much safer and cheaper than in the real world. However, due to discrepancies between the simulated and real data, simulator-trained policies often fail when transferred to real robots. One common approach to bridge the visual sim-to-real domain gap is domain randomization (DR). While previous work mainly evaluates DR for disembodied tasks, such as pose estimation and object detection, here we systematically explore visual domain randomization methods and benchmark them on a rich set of challenging robotic manipulation tasks. In particular, we propose an off-line proxy task of cube localization to select DR parameters for texture randomization, lighting randomization, variations of object colors and camera parameters. Notably, we demonstrate that DR parameters have similar impact on our off-line proxy task and on-line policies. We, hence, use off-line optimized DR parameters to train visuomotor policies in simulation and directly apply such policies to a real robot. Our approach achieves 93% success rate on average when tested on a diverse set of challenging manipulation tasks. Moreover, we evaluate the robustness of policies to visual variations in real scenes and show that our simulator-trained policies outperform policies learned using real but limited data. Code, simulation environment, real robot datasets and trained models are available at https://www.di.ens.fr/willow/research/robust_s2r/.

Guillaume Le Moing, Jean Ponce, Cordelia Schmid

This paper presents WALDO (WArping Layer-Decomposed Objects), a novel approach to the prediction of future video frames from past ones. Individual images are decomposed into multiple layers combining object masks and a small set of control points. The layer structure is shared across all frames in each video to build dense inter-frame connections. Complex scene motions are modeled by combining parametric geometric transformations associated with individual layers, and video synthesis is broken down into discovering the layers associated with past frames, predicting the corresponding transformations for upcoming ones and warping the associated object regions accordingly, and filling in the remaining image parts. Extensive experiments on multiple benchmarks including urban videos (Cityscapes and KITTI) and videos featuring nonrigid motions (UCF-Sports and H3.6M), show that our method consistently outperforms the state of the art by a significant margin in every case. Code, pretrained models, and video samples synthesized by our approach can be found in the project webpage https://16lemoing.github.io/waldo.

Quentin Le Lidec, Wilson Jallet, Ivan Laptev et al.

Reinforcement learning (RL) and trajectory optimization (TO) present strong complementary advantages. On one hand, RL approaches are able to learn global control policies directly from data, but generally require large sample sizes to properly converge towards feasible policies. On the other hand, TO methods are able to exploit gradient-based information extracted from simulators to quickly converge towards a locally optimal control trajectory which is only valid within the vicinity of the solution. Over the past decade, several approaches have aimed to adequately combine the two classes of methods in order to obtain the best of both worlds. Following on from this line of research, we propose several improvements on top of these approaches to learn global control policies quicker, notably by leveraging sensitivity information stemming from TO methods via Sobolev learning, and augmented Lagrangian techniques to enforce the consensus between TO and policy learning. We evaluate the benefits of these improvements on various classical tasks in robotics through comparison with existing approaches in the literature.

Chen Sun, Calvin Luo, Xingyi Zhou et al.

We aim to investigate whether end-to-end learning of visual reasoning can be achieved with general-purpose neural networks, with the help of visual pretraining. A positive result would refute the common belief that explicit visual abstraction (e.g. object detection) is essential for compositional generalization on visual reasoning, and confirm the feasibility of a neural network "generalist" to solve visual recognition and reasoning tasks. We propose a simple and general self-supervised framework which "compresses" each video frame into a small set of tokens with a transformer network, and reconstructs the remaining frames based on the compressed temporal context. To minimize the reconstruction loss, the network must learn a compact representation for each image, as well as capture temporal dynamics and object permanence from temporal context. We perform evaluation on two visual reasoning benchmarks, CATER and ACRE. We observe that pretraining is essential to achieve compositional generalization for end-to-end visual reasoning. Our proposed framework outperforms traditional supervised pretraining, including image classification and explicit object detection, by large margins.

Thomas Chabal, Robin Strudel, Etienne Arlaud et al.

This paper addresses the problem of copying an unknown assembly of primitives with known shape and appearance using information extracted from a single photograph by an off-the-shelf procedure for object detection and pose estimation. The proposed algorithm uses a simple combination of physical stability constraints, convex optimization and Monte Carlo tree search to plan assemblies as sequences of pick-and-place operations represented by STRIPS operators. It is efficient and, most importantly, robust to the errors in object detection and pose estimation unavoidable in any real robotic system. The proposed approach is demonstrated with thorough experiments on a UR5 manipulator.

Zerui Chen, Rolandos Alexandros Potamias, Shizhe Chen et al.

Generating realistic 3D hand-object interactions (HOI) is a fundamental challenge in computer vision and robotics, requiring both temporal coherence and high-fidelity physical plausibility. Existing methods remain limited in their ability to learn expressive motion representations for generation and perform temporal reasoning. In this paper, we present HO-Flow, a framework for synthesizing realistic hand-object motion sequences from texts and canoncial 3D objects. HO-Flow first employs an interaction-aware variational autoencoder to encode sequences of hand and object motions into a unified latent manifold by incorporating hand and object kinematics, enabling the representation to capture rich interaction dynamics. It then leverages a masked flow matching model that combines auto-regressive temporal reasoning with continuous latent generation, improving temporal coherence. To further enhance generalization, HO-Flow predicts object motions relative to the initial frame, enabling effective pre-training on large-scale synthetic data. Experiments on the GRAB, OakInk, and DexYCB benchmarks demonstrate that HO-Flow achieves state-of-the-art performance in both physical plausibility and motion diversity for interaction motion synthesis.

Paul Hongsuck Seo, Arsha Nagrani, Cordelia Schmid

In this report, we describe our submission to the Ego4D AudioVisual (AV) Speech Transcription Challenge 2022. Our pipeline is based on AVATAR, a state of the art encoder-decoder model for AV-ASR that performs early fusion of spectrograms and RGB images. We describe the datasets, experimental settings and ablations. Our final method achieves a WER of 68.40 on the challenge test set, outperforming the baseline by 43.7%, and winning the challenge.

Arsha Nagrani, Jasper Uijilings, Shyamal Buch et al.

Video reasoning models are a core component of egocentric and embodied agents. However, standard benchmarks for assessing models provide only evaluation of the output (e.g. the answer to a question), without evaluation of intermediate reasoning steps, and most provide answers only in the text domain. We introduce Minerva-Ego, a benchmark for evaluating complex egocentric visual reasoning. We extend recent high-quality video data sources recorded from egocentric / embodied settings with a set of challenging, multi-step multimodal questions and spatiotemporally-dense human-annotated reasoning traces. Benchmarking experiments show that state-of-the-art models still have a large gap to human performance. To investigate this gap in detail, we annotate each reasoning trace in the dataset with the objects of interest required to solve the question, as spatiotemporal mask annotations. Through extensive evaluations, we identify that prompting frontier models with hints of 'where' and 'when' to look yields substantial improvements in performance. Minerva-Ego can be downloaded at https://github.com/google-deepmind/neptune.

Matthieu Futeral, Cordelia Schmid, Benoît Sagot et al.

Current multimodal machine translation (MMT) systems rely on fully supervised data (i.e models are trained on sentences with their translations and accompanying images). However, this type of data is costly to collect, limiting the extension of MMT to other language pairs for which such data does not exist. In this work, we propose a method to bypass the need for fully supervised data to train MMT systems, using multimodal English data only. Our method, called ZeroMMT, consists in adapting a strong text-only machine translation (MT) model by training it on a mixture of two objectives: visually conditioned masked language modelling and the Kullback-Leibler divergence between the original and new MMT outputs. We evaluate on standard MMT benchmarks and the recently released CoMMuTE, a contrastive benchmark aiming to evaluate how well models use images to disambiguate English sentences. We obtain disambiguation performance close to state-of-the-art MMT models trained additionally on fully supervised examples. To prove that our method generalizes to languages with no fully supervised training data available, we extend the CoMMuTE evaluation dataset to three new languages: Arabic, Russian and Chinese. We further show that we can control the trade-off between disambiguation capabilities and translation fidelity at inference time using classifier-free guidance and without any additional data. Our code, data and trained models are publicly accessible.

Xingyi Zhou, Anurag Arnab, Shyamal Buch et al.

An ideal model for dense video captioning -- predicting captions localized temporally in a video -- should be able to handle long input videos, predict rich, detailed textual descriptions, and be able to produce outputs before processing the entire video. Current state-of-the-art models, however, process a fixed number of downsampled frames, and make a single full prediction after seeing the whole video. We propose a streaming dense video captioning model that consists of two novel components: First, we propose a new memory module, based on clustering incoming tokens, which can handle arbitrarily long videos as the memory is of a fixed size. Second, we develop a streaming decoding algorithm that enables our model to make predictions before the entire video has been processed. Our model achieves this streaming ability, and significantly improves the state-of-the-art on three dense video captioning benchmarks: ActivityNet, YouCook2 and ViTT. Our code is released at https://github.com/google-research/scenic.